tensorflow/tflite-micro, also known as TensorFlow Lite for Microcontrollers (TFLM), is a C++ project designed to run deep learning models on memory-constrained bare-metal systems, DSPs, and microcontrollers. It uses the Apache-2.0 license, and the provided record contains no latest release version, so it cannot be introduced as the changes of a specific version.
What this project targets is model execution on embedded systems with tens to hundreds of kilobytes of memory. It targets models converted through LiteRT and TensorFlow, and is described as a TinyML runtime that drives deep learning on devices with limited memory and compute resources. The architectures tested in CI include Arm Cortex-M, Cadence Xtensa, RISC-V, Qualcomm Hexagon, and x86/Windows. So rather than being an implementation limited to a single type of MCU, it is a project that supports a range of embedded and development environments.
That said, this record does not present performance benchmarks, per-model memory usage, or accuracy changes. The memory scale and architecture list come from the README description and are not independent measurements. Which models work with how much memory in an actual deployment cannot be judged from this material alone, and the release status is likewise unverified.